Interesting benchmarks. I'm surprised vLLM is performing so poorly given all the corporate investment/attention it's getting. Is it because the contributions are more distributed than the average LLM team?
NASA's report server is a national treasure, especially the material from the 50s and 60s that he references. Some of the most crisp and succinct technical writing you'll find, and you can infer a lot about how they ran projects. Declassified NRO reports are also very good - you can see the Lockheed Skunk Works principles in action. Example: https://www.nro.gov/Portals/135/documents/foia/declass/WS117...
You could try a battery powered phone charger since it's a "relatively simple" first project. The big hurdle for learning these types of tools is usually "What buttons do I press to create the output that I want.
For the electrical side, there are plenty of schematics online that you can try to copy or use as a starting point. And the CAD side can be a simple box with snap fits. I'd recommend OnShape if you're just starting out since it's the lowest barrier of entry, but Fusion 360 is also good. All in, it should be <$150 for the PCBs + Components + 3D Prints.
After you get the satisfaction of seeing your device charge from something you made, then you'll start getting the itch and find more excuses to make things.
This strikes me as something that will fade over time, though. We will eventually learn to recognize fake empathy, just as once upon a time when a corporation said "Your business is important to us and we're trying to get a support person on the line for you as quickly as possible", it was believable and there was a good chance your customer believed it. Now of course we've all got a pretty good idea it's not true.
An AI can not empathize. We don't even really want it to; who wants to build an AI that "really" experiences losing a limb or losing a daughter? Not anyone I want actually building AIs. So this isn't even about whether they're "really conscious" or any of those somewhat tedious debates; even if they are human-level AI already they literally can't empathize. See the recent article where Meta's overly helpful AI yielded an answer as to how New York's public schools treated its disabled child. Even if the text was completely accurate it still had no standing to emit such text.
It's... just barely probable with two people. Assuming 30 whistleblowers, and an actuarial likelihood of 1.2% death[1] in a given year at their ages, we can use binomial probability[2] to determine the chance of two dying in a given year.
The math works out to a 4.4% chance. That's a bit of an eye raiser, but still just believable. (Look, we all believe it can happen we roll a critical in D&D ;)
3 or more dying in a year, we're at 0.5% - I'd really assume that's deliberate.
But it's okay to continue spamming postings that don't get updated for a year, wasting users' time on here? This is YCombinator forums right and these companies have to look good for VCs I suppose.
Super promising & interesting, but also not really an advanced in throughout vs today's switches. That's 16x 800Gbps. Awesome & boggling, but for something that'll be coming out a good bit down the road.
Just for reference, a MI300X is good for 5.3TBps aka 42Tbps.
Overally my hope still is we see some interesting disaggregated architectures form. Rather than be used for ToR switches only, I hope this sort of thing can be used for closely connecting small clusters. Ideally an advanced flash controller also would have such a device.
The material cost of cables and their energy efficiency grow terrifying. Photonics remain a key to helping us physically scale our systems effectively.
Anything that uses their servers they could potentially charge for. Of course they must also make it possible to implement these things yourself without using their servers.
> A popular venture studio based out of NYC is like this
Sounds like Fractal. Supposedly, the value add is that they've already done the due diligence and market research on some product idea. They match a team (CEO + CTO) to the idea and provide the funding. Do not know of any well known companies to have come out of this model, but if they're still around, it must be generating some returns for it to be worthwhile.
Yes they would. It would be a significant and catastrophic mistake to restrict trading based on feelings of envy. It would be technically negative in every measurement possible.
So many poor policies originate from Envy and poor reasoning not grounded in logic and understanding of free markets and economics. This would be yet another classic example of that.
I'd be very surprised if organizations such as the CIA or similar haven't at least put some resources toward researching the viability of weaponizing MRSA or similar.
While obviously horrific, there's a large difference between paying a bunch of paramilitary members of a country where out and out murder by government is already an issue, and murdering citizens using arcane secretive methods such as suicide by hanging and magic mrsa infection.
If boeing is SO important to the government that justice doesn't matter, then it's probably a fuckload easier to bribe a judge or two and make sure the court cases never get anywhere than to orchestrate not one, but two, clandestine hits.
I like CL’s solution to constructors which is basically “specialize this generic function (SHARED-INITIALIZE or INITIALIZE-INSTANCE) with an :AFTER method”. You reliably run all the initialization code for each class involved and you don’t have to remember to call CALL-NEXT-METHOD (CL’s spelling of super)
Edit: I see that post refers to Dylan, which is more like CL than python in the important ways. IMO, sleeping on CL’s object system CLOS was a huge mistake of the “Java/C++ era” of our industry.
You are in Denmark? WOW!! From here in the US, my impression of Denmark is a small country of really smart people with gorgeous women having really good lives!!! Your post questions that impression ....
Here in the US, there are what seem to be related issues. Opinions about what is wrong and why differ. It can seem that we don't address the issues or what is wrong, for example, it can appear that the media believes that making clear the issues and faults are too deep for the audience of the media. For example, so far I've yet to see good engineering style graphs of even the basic measures from the economy, health, education, happiness, etc. In simple terms, the birth rate is so low we are going extinct -- some graphs over time might provide some insight. To avoid going extinct should be worth some graphs of related data?
Here is one guess: Some people want more government involvement, activity, services, ..., at the national level. Involved are money and power, money collected from taxes and/or borrowed and then spent and the people spending the money getting power. Then for the money the guess is that the spending is too inefficient, that is, too often wasted. For the power, it can stop what might be solutions. So, with the waste, we are throwing away our efforts, and with the power we are blocked from better approaches.
For this view, the US has some old images, from history, the movies, some parts of education, and even from some of recent history: So, the old images are from, say, the movie The Big Sky from the 1830s of 20 or so men and one Native Indian woman leaving Saint Louis on the Mississippi River (major US river from the north east going south and dumping into the Gulf of Mexico) and the Missouri River (from the northwest of the US close to the Pacific Ocean and flowing ~2000 miles to the Mississippi) to trade with the Blackfoot tribe to give them woven blankets, guns, tools and get beaver pelts.
The whole effort was lawless, that is, no government, laws, police, political power, regulations. So there was fighting, killing. No medical care, so there was suffering. There were lots of injuries from struggles with the river. It was all highly risky with lots of struggles and losses but in the end successful.
There were more such movies about the US, people on their own, risking everything, fighting, struggling, having failures, in the end being successful. There was Henry Ford, the Wright Brothers (had a bicycle shop), etc.
To the present, there were two guys at Stanford University in a project for indexing the contents of libraries. They thought of indexing the Web sites on the Internet and were offered ~$1 million for their work and Web site. They turned down the $1 million, started a company, and had success. It was all risky. It's now worth ~$1 trillion, called Google. Other successes, risky, one or a few founders, include Microsoft, Walmart, FedEx, Facebook, Amazon, Burger King, McDonald's. One guy, sole, solo founder, used some old Dell computers, an early version of Microsoft's Windows and Web software ASP.NET, built a romantic matchmaking service, grew it, ..., and sold out for ~$550 million.
Mostly these successes are examples of significant increases in economic productivity, that is, how much utility get from each hour of work. Personal computers? Killed off the typewriters with just astounding increases in productivity; same for email; Zoom, .... In cars, using digital electronics to get fuel mixture and spark timing a lot better resulted in the engines lasting much longer -- if looked carefully into what was going on with the old carburetors and breaker point ignitions and the huge ways they were wrong (unburned gasoline diluted the engine oil and, thus, wore out the main engine parts) can understand. Tires -- progress in the synthetic rubber chemistry made the tires last much longer, maybe 5 times as long.
The theme is, as in the movie, leave individuals with a minimum of taxes and government power and let them try and take the risks. The victories can generate massive increases in productivity, wealth, and a better life.
Heck, I like music, A. Vivaldi, J. Bach, ..., S. Barber. Used to be, had to go to live performances. Then we got audio tapes, vinyl records, CDs, DVDs, and the Internet. So, now just from a few keystrokes I can get terrific performances of Vivaldi's The Four Seasons, Bach's Chaconne, ... Barber's Adagio for Strings.
Those are some of the ideas that have floated around. Money and power: The idea that just these two are the main causes of all of society's problems is tough to accept. But, for the problems in Denmark, maybe these two should get some attention.
That's where the economics question comes in. How do you make a living as a farmer selling a very specific variety of squash, when there are now (let's say) 9000 varieties of squash on the market, and the same number of people out there buying squash?
Octopus V4 is a graph of language models that creates a network of specialized open-source AI language models, where it acts as the master node coordinating with various domain-specific AI models as worker nodes. It aims to reduce the high computational, data, and energy demands that limit current large language model development and addresses restrictions imposed by proprietary models. In testing, such as the MMLU benchmark, Octopus V4 has achieved scores that suggest it can compete with and potentially exceed proprietary models' performance, thus showing potential to match or surpass proprietary models through a collaborative, open-source approach.
You're right. We also have proverbs like "two heads are better than one" and "standing on the shoulder of giants" so people recognise that both sides are important and have their value.
Right now, I'm working on my own on a personal project attempting to do something a little novel and I appreciate being able to go back and refine my ideas/previous code based on things I learn and additional thinking (even rewriting from scratch), when I'm more likely to face friction (like "stick to the suboptimal approach; it's not that bad") and cause trouble for teammates if I was working with someone else. So the value of working alone speaks more to me currently than the value of working in teams, but they both have their place.
